Five failures the architecture cannot reach, stated as plainly as the six it can.

Why the boundary matters commercially

The buyer most likely to be disappointed is the one who was never told where it ends.

Six named problems, each with an owning module, is a strong claim, and strong claims invite generous extrapolation. A reader whose actual difficulty is that no positive-expectancy method exists yet will read the six and conclude the system addresses their situation, because their situation certainly involves losing money and the six certainly involve losing money. It does not. Stating the boundary is not a disclaimer bolted onto the end of the argument; it is the argument's precondition, because a claim with no stated edge is not a specification and cannot be checked by anyone.

The first limit

Governing a method with no edge produces a well-documented loss.

The architecture measures, sizes, routes and audits an existing method. It does not manufacture expectancy, and applied to a method whose true expectancy is negative it will do precisely what it is built to do: report the negative expectancy accurately, compress deployment as drawdown accumulates, and eventually lock the account. That is a materially better outcome than the alternative, and it is worth roughly what it costs — the loss arrives smaller, slower and correctly labelled. It is not, however, the outcome most buyers picture, and the difference between a governed loss and a rescued account should not be blurred. The one genuine consolation is that the diagnosis arrives with capital still in the account, which is the precondition for the method being replaced rather than the trader being finished.

The second and third limits

Nothing in the stack chooses what to trade, or supplies capital.

There is no instrument anywhere in the eleven workbooks that selects an instrument, a session, or a setup. Volatility tooling assesses whether conditions support the intended management model once an asset is already in play; it does not rank candidates. Selection stays entirely with the operator, and a persistent selection error will show up only as degraded expectancy months later, attributed to no particular cause. Capitalisation is the same shape of limit: gates preserve capital and cannot create it, and an account too small for its intended risk-per-trade faces an arithmetic problem that no governance layer converts into a survivable one. Gates will compress deployment faithfully as that account declines, which is correct behaviour and no remedy at all.

The fourth limit

Entry quality is observable only through outcomes, and only slowly.

The MAE/MFE Lab measures a great deal about execution — adverse excursion, favourable excursion, capture efficiency, giveback, stop efficiency, fee drag — and none of it answers whether the setup should have been taken. It answers whether the opportunity that appeared was converted. A trader entering consistently poor setups and managing them impeccably will produce clean efficiency readings and weak expectancy, and the instruments will report both truthfully without connecting them. The connection requires the operator to form a hypothesis the machinery cannot form, which is the standing consequence of a system built with no entry tooling.

• A governed negative edge produces a smaller, slower, correctly-labelled loss.
• Selection and capitalisation sit wholly outside the instruments.
• Efficiency readings can be excellent while setup quality is the actual problem.

The fifth limit

The machine can produce the finding; it cannot produce the response.

The final limit is the one no architecture escapes. Every module in the stack terminates in a reading, and a reading changes nothing until someone acts on it. Adherence counts can be accurate and ignored; a compressed tier can be authorised and overridden, with the override logged and then repeated. The logging is real and the audit trail is genuine, and neither constitutes enforcement. The system's authority is entirely delegated by the operator and revocable by them at any moment, which means the last and largest failure mode is not measurable by anything inside it.

The key idea

A stated boundary is worth more than an implied one.

Every limit here is structural rather than provisional — none of them is a feature scheduled for a later edition, and none would be resolved by more workbooks. They follow from what the architecture is: an evidence-and-authority layer over a method the operator supplies, chooses, funds and ultimately obeys or does not. Read alongside the six problems it does solve, the effect is to narrow the claim considerably. That is the intention. A buyer who knows exactly which of their problems is being addressed can judge whether it is the expensive one.